| 引用本文: | 王继东,张迪.基于侧输出融合卷积神经网络的电能质量扰动分类方法[J].电力自动化设备,2021,41(11): |
| WANG Jidong,ZHANG Di.Power quality disturbance classification method based on side-output fusion convolutional neural network[J].Electric Power Automation Equipment,2021,41(11): |
|
| 摘要: |
| 针对传统电能质量扰动分类方法分类准确率低、人工选择特征困难等缺点，提出了一种基于深度学习的侧输出融合卷积神经网络用于电能质量扰动信号分类。首先，对电能质量扰动信号进行预处理，使输入信号数据标准化，有利于提升所提方法的收敛速度和精度。在传统卷积神经网络中引入侧输出融合结构，通过组合卷积低、中和高层的信息进行特征融合，以更好地对输入信号进行分类。针对实测数据不足和信号数据类型分布不均衡等问题，采用数据增强的方法对信号进行处理。仿真和实测数据验证表明，所提方法可以自动进行特征提取和优化，具有分类速度快、分类准确率高等优点。 |
| 关键词: 电能质量 扰动分类 侧输出融合卷积神经网络 深度学习 特征提取 |
| DOI：10.16081/j.epae.202107010 |
| 分类号:TM732 |
| 基金项目:国家重点研发计划项目(2016YFB0901102) |
|
| Power quality disturbance classification method based on side-output fusion convolutional neural network |
|
WANG Jidong, ZHANG Di
|
|
Key Laboratory of Smart Grid of Ministry of Education, Tianjin University, Tianjin 300072, China
|
| Abstract: |
| Aiming at the disadvantages of the traditional power quality disturbance classification methods, such as low classification accuracy and difficulty in manually selecting features, a SFCNN(Side-output Fusion Convolutional Neural Network) based on deep learning is proposed for power quality disturbance classification. Firstly, the power quality disturbance signal is preprocessed to standardize the input signal data, which is beneficial to improve the convergence speed and accuracy of the proposed method. Then, the side-output fusion structure is introduced into the traditional convolutional neural network, and feature fusion is carried out by combining the information of the low, middle and high layers of convolution to better classify and recognize the input signal. Aiming at the problems of insufficient measured data and unbalanced distribution of signal data types, the data enhancement method is used to process the signal. Simulation and actual data verification show that the proposed method can automatically perform feature extraction and optimization, and has the advantages of fast classification speed and high classification accuracy. |
| Key words: power quality disturbance classification side-output fusion convolutional neural network deep learning feature extraction |
